Prevention of necrotizing enterocolitis with probiotics: a systematic review and meta-analysis

Rationale

Necrotizing enterocolitis (NEC) is a gastrointestinal (GI) syndrome characterized by transmural inflammation and necrosis of the large or small bowel and subsequent translocation of gas-forming organisms into the intestinal wall (Thompson & Bizzarro, 2008; Morgan, Young & McGuire, 2011). Primarily seen in infants, the incidence of NEC is inversely correlated with gestational age (GA) and birth weight (Sharma & Hudak, 2013; Choi, 2014). The overall incidence of NEC in all infants ≤33 weeks GA in a survey of Canadian neonatal intensive care units (NICUs) was 5%, and 7% for infants less than 1500 g birth weight in 2013 (Henry & Moss, 2008).

The consequences of NEC are potentially devastating—20% to 40% of patients require surgical intervention and mortality ranges from 15% to 30% (Holman et al., 2006; Sia et al., 2014). Survivors of NEC risk significant morbidity including short gut syndrome, strictures, and neurodevelopmental impairment (Henry & Moss, 2008; Holman et al., 2006).

The signs and symptoms of NEC were classified by Bell in 1978 and gave rise to modified criteria for diagnosis of NEC in 1996 by Neu (Sia et al., 2014). The modified Bell’s criteria describe the systemic clinical signs of NEC, the important GI signs (which can help differentiate NEC from sepsis), and the radiologic features.

The immature GI tract of preterm infants is particularly susceptible to mucosal injury from a variety of factors. Intestinal and immunological deficiencies associated with prematurity, enteral feeding, microbial overgrowth, and circulatory instability have all been implicated in the pathogenesis of NEC (Schanler, 2015).

Recent research has focused on microbial overgrowth in the GI tract of premature infants, with an overabundance of pathogenic organisms and lack of microbial diversity being key discoveries. These observations imply that a disturbance in the microbiome, and not a single pathogen, may be a causative factor of NEC (Terrin, Scipione & De Curtis, 2014). The lower prevalence of protective Lactobacillus or Bifidobacterium species in preterm infants compared to term infants make probiotics a potential intervention for the prevention of NEC (Morowitz et al., 2010).

Previous systematic reviews

At the time of our search, there were two recent systematic reviews and meta-analyses on this topic (AlFaleh, 2014; Yang et al., 2014). The Cochrane review on this topic is thorough, but it was last updated in October 2013 (AlFaleh, 2014). The Yang (Terrin, Scipione & De Curtis, 2014) review included many of the same studies but included additional studies as a result of a Chinese trial database search (Ke, Su & Li, 2008; Huang, Yang & Huang, 2009; Ren, 2010; Di & Li, 2010). Since the publication of these two systematic reviews, more large randomized clinical trials have been published.

Objective

The objective of this systematic review was to assess the efficacy and safety of probiotics for the prevention of NEC in premature infants. We planned to update the previous systematic reviews (AlFaleh, 2014; Yang et al., 2014) using similar eligibility criteria.

Protocol/registration

The systematic review methods and analysis plans were undertaken according to published guidelines by PRISMA (Moher et al., 2009).

Eligibility criteria

Studies: All randomized clinical trials were considered for inclusion. No language restrictions were applied.

Participants: Infants of less than 37 weeks gestation or weighing less than 2,500 g at birth.

Interventions: Probiotics in any species and any dose, or prebiotic/probiotic combinations (synbiotics) of any species and any dose.

Comparators: Probiotic products with different species than the intervention group (i.e., RCTs comparing one species to another head-to-head), placebo, or standard therapy.

Outcomes: The primary outcome of the review was the incidence of severe NEC (Bell’s Stage 2 or greater). Secondary outcomes included all-cause mortality, all-cause sepsis, culture-proven sepsis, bacterial sepsis, fungal sepsis, length of stay in hospital, time to achieve full feeds, duration of parenteral nutrition, and weight gain.

Information sources & search

Study selection

After de-duplication, two reviewers independently screened titles and abstracts for inclusion using a standardized screening tool. Full text screening was completed independently in duplicate by two authors using a full-text screening tool. Cohen’s Kappa was used to assess agreement between the two reviewers on the selection of full-text articles for inclusion (Freelon, 2015).

Data collection process & data Items

A standardized data collection form was developed a priori and two authors independently extracted the relevant outcomes and validity criteria from the new trials. The data pertaining to the old trials, including risk of bias assessment, was extracted by one author. Disagreements were resolved by consensus and a third party was consulted if necessary. Author contact was attempted for outcome data in the included trials which was missing or unclear. The complete list of the data extracted from the included trials is included in Appendix B.

Study outcome data published in duplicate was included once, but all versions of the publication were utilized for maximal data extraction. In the event of inconsistency between multiple reports of one study, the peer-reviewed publication was used as the primary data set.

Risk of bias within and across studies

Two authors independently assessed the risk of bias for each of the new included studies using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins, Deeks & Altman, 2011). A summary table and a graph for risk of bias were created using Review Manager (RevMan) software (The Cochrane Collaboration, 2014). The risk of bias asessments from the studies included in the previous systematic reviews (AlFaleh, 2014; Yang et al., 2014) were double-checked for accuracy by a single author.

Synthesis of results

When possible, the results were synthesized using RevMan 5.3 (The Cochrane Collaboration, 2014). A random effects model (Higgins, Deeks & Altman, 2011) was chosen to account for the clinical and statistical heterogeneity expected when including different species and regimens of probiotics, different neonatal ages and weights, different feeding regimens (breast milk, formula, combination feeding, and parenteral nutrition supplementation as needed), as well as the varied countries conducting RCTs in this area. Relative risks (RRs) with 95% confidence intervals (CIs) were used for dichotomous variables and mean differences (MDs) with 95% CIs for continuous variables. If the continuous variables in the studies were measured in different scales, we calculated the standardized mean difference (SMD).

Analysis was done on an intention-to-treat (ITT) basis (Higgins, Deeks & Altman, 2011). If patients discontinued the intervention after randomization, they were still counted in our analysis for outcomes (such as mortality) where this was possible. Author contact was attempted to clarify any missing outcome data.

If trials had two intervention arms, both of which contained a probiotic, both probiotic arms were included and the number of patients in the comparator arm divided by the number of active arms to prevent double counting. If the trial had two or more intervention arms and only one of them contained a probiotic, the data from the corresponding non-probiotic arm was used as the comparator (Higgins, Deeks & Altman, 2011). In trials where patients received a co-intervention, the co-intervention had to be present in both the active and control arms to be included.

The I² statistic was used to quantify statistical heterogeneity (the percentage of total variation across studies due to heterogeneity). Statistical heterogeneity as measured by I² was described as “small” (≤25%), “moderate” (between 26% and 49%) and “large” (≥50%) (Higgins et al., 2003). Forest plots were visually inspected for possible sources of heterogeneity.

Additional analysis

Subgroup analysis was planned a priori for the following subgroups: infant weight (extremely-low birth weight (ELBW) (less than 1,000 g) and very low birth weight (VLBW) (less than 1,500 g)), timing of probiotic initiation, duration of probiotic therapy, sepsis types (including “any sepsis”), and use of breastmilk vs. formula for feeding.

Study selection

The previously published systematic reviews included a total of 37 unique randomized clinical trials (Ke, Su & Li, 2008; Huang, Yang & Huang, 2009; Ren, 2010; Yang, Yi & Gan, 2011; Romeo et al., 2011; Underwood et al., 2009; Al-Hosni et al., 2012; Bin-Nun et al., 2005; Braga et al., 2011; Costalos et al., 2003; Dani et al., 2002; Demirel et al., 2013; Fernandez-Carrocera et al., 2013; Kitajima et al., 1997; Li et al., 2004; Lin et al., 2005; Lin et al., 2008; Manzoni et al., 2006; Manzoni et al., 2009; Mihatsch et al., 2010; Millar et al., 1993; Mohan et al., 2006; Jacobs et al., 2013; Reuman et al., 1986; Rojas et al., 2012; Rougé et al., 2009; Samanta et al., 2009; Sari et al., 2011; Stratiki et al., 2007). Electronic database searches (including the 2016 search update) yielded 475 citations, conference searching yielded 115 citations, and clinical trials database searching yielded 35 citations. After de-duplication, 412 citations remained for title and abstract screening (see Fig. 1 for the detailed flow diagram of study selection). Cohen’s Kappa was 0.723 (good agreement) between the two reviewers for selection of new full-text trials for inclusion (Altman, 2001). The study by Manzoni et al. (2014) was included in our review as it was a randomized extension of a previously published trial (Manzoni et al., 2009). The ProPrems study was added to the previous review as unpublished data, but is now published and was included in our review (Jacobs et al., 2013). The updated search in 2016 resulted in a follow-up to the Oncel trial (Oncel et al., 2014) (Akar; Akar et al., 2016), and new trials by Costeloe (Costeloe et al., 2016; previously on our ongoing trials list), Dilli (Dilli et al., 2015), Dutta (Dutta, Ray & Narang, 2015), Sinha (Sinha et al., 2015) and Tewari (Tewari, Dubey & Gupta, 2015). Three trials included in previous reviews were excluded from our review as they were determined to be non-randomized (Fu & Song, 2012; Hunter et al., 2012; Li, Qiao & Huang, 2011). The overall updated search added a total of 13 randomized controlled trials (two trials split due to multiple arms) with over 5,000 new evaluable patients to previous systematic reviews, bringing the total to 42 included trials (Hua, Tang & Mu, 2014; Manzoni et al., 2014; Oncel et al., 2014; Costeloe et al., 2016; Dilli et al., 2015; Dutta, Ray & Narang, 2015; Sinha et al., 2015; Tewari, Dubey & Gupta, 2015; Patole et al., 2014; Roy et al., 2014; Serce et al., 2013; Van Niekerk et al., 2014a).

Study Characteristics

When verifying the outcome data included in the previous reviews, a number of methodological flaws and errors of data synthesis were noted (AlFaleh, 2014; Yang et al., 2014). A decision was made to re-extract the data from the “old trials” instead of re-entering the data from the previously published reviews (see Appendix C).

See Table 1 for characteristics of included studies. All studies were conducted in preterm infants admitted to the NICU. Twenty-four studies limited birth weight to 1,500 g or less. Weight was not part of the inclusion criteria in nine studies (Ke, Su & Li, 2008; Hua, Tang & Mu, 2014; Costalos et al., 2003; Millar et al., 1993; Mohan et al., 2006; Stratiki et al., 2007; Costeloe et al., 2016; Dutta, Ray & Narang, 2015; Tewari, Dubey & Gupta, 2015). Gestational age was not part of the inclusion criteria in five studies but all of these studies had birth weight inclusion criteria for preterm infants less than 1,500 g (Braga et al., 2011; Kitajima et al., 1997; Lin et al., 2005; Manzoni et al., 2006; Manzoni et al., 2014). One trial did not specify gestational age but enrolled babies 1,500–2,500 g (Sinha et al., 2015). Five trials were translated from Chinese for use in the review (Ke, Su & Li, 2008; Huang, Yang & Huang, 2009; Ren, 2010; Hua, Tang & Mu, 2014; Li et al., 2004).

Type of feeding was variable across the included trials. Nine trials included infants exclusively fed breastmilk (Lin et al., 2005; Manzoni et al., 2006; Rougé et al., 2009; Samanta et al., 2009; Sinha et al., 2015; Tewari, Dubey & Gupta, 2015; Roy et al., 2014; Van Niekerk et al., 2014b). One trial had infants fed exclusively preterm formula (Stratiki et al., 2007). The trials published in Chinese did not consistently specify this information on translation (Ke, Su & Li, 2008; Huang, Yang & Huang, 2009; Ren, 2010; Yang, Yi & Gan, 2011; Li et al., 2004). Costeloe et al. (2016) had 46% of infants exclusively fed breastmilk, but the rest of the infants had a combination of feeding types.

Overall, the number of trials were split evenly between multiple species and single species probiotics (22 trials each). The Sari trial (Bacillus coagulans formerly known as Lactobacillus sporogenes) (Sari et al., 2011) and the Tewari trial (Bacillus clausii) (Tewari, Dubey & Gupta, 2015) used single species that were not used in any other trial. Sinha used a multi-organism product containing eight species (Sinha et al., 2015). All studies used a variety of organisms and dose regimens. Comparators were matching placebo, standard therapy, or prebiotics (two trials) (Manzoni et al., 2014; Dilli et al., 2015). There were no trials comparing one probiotic preparation with another, but two trials had multiple arms with different probiotics (Romeo et al., 2011; Underwood et al., 2009). One trial used varying durations of probiotics and doses but fit within the range of doses and duration of therapy seen with all included trials, so the three treatment arms were combined into one (Dutta, Ray & Narang, 2015).

Timing of probiotic initiation was variable. Twenty-one trials started probiotics with the first feed, six trials started within 48 h of birth, one within 72 h, four within the first week, and in twelve trials therapy started at the “more than 48 h” time point.

Duration of probiotic therapy ranged from seven days to six weeks. One trial did not specify a duration of therapy (Stratiki et al., 2007). Most studies were classified in the “28 days or more” subgroup for the purposes of analysis by extraction of the actual duration of therapy (when provided) in trials that specified duration as “until discharge.”

All infants

The primary outcome, severe NEC, was significantly reduced in infants who received probiotics compared to placebo with 38 trials (10,520 patients) reporting on this outcome—RR 0.53 95% CI [0.42–0.66]—see Fig. 2. The incidence of culture-proven sepsis was not different between the probiotics and control—RR 0.88 95% CI [0.77–1.00] in 31 trials comprising 8,707 patients, see Fig. 3. The incidence of all-cause mortality was significantly reduced in infants receiving probiotics in 29 trials (9,507 patients)—RR 0.79 95% CI [0.68–0.93] (Fig. 4). Other statistically significant findings included shorter duration of hospitalization, increased weight gain (g/day), and reduced time to reach full enteral feeds, all in favor of using probiotics (Table 2).

There was a moderate to large degree of heterogeneity in the results for culture-proven sepsis, duration of hospitalization, duration of parenteral nutrition, and time to achieve full feeds.

Table 2:

Additional important findings.

Outcome	Number of studies / participants	Effect size	95% CI	I2 (%)
All Infants
Bacterial sepsis	9 / 2212	RR 0.86	0.62 to 1.18	52
Fungal sepsis	12 / 3756	RR 0.67	0.43 to 1.06	10
Duration of hospitalization (days)	16 / 4915	MD −3.2	−5.5 to −0.9	84
Weight gain (g/day)	3 / 314	MD +1.7	1.0 to 2.3	0
Time to achieve full feeds (days)	17 / 4448	MD −1.2	−2.2 to −0.1	93
VLBW infants
Culture-proven sepsis	24 / 6616	RR 0.93	0.82 to 1.05	15
Duration parenteral nutrition (days)	4 / 1210	MD −1.2	−2.3 to −0.02
ELBW infants
Culture-proven sepsis	6 / 1703	RR 0.95	0.72 to 1.26	41
Mortality	4 / 1122	RR 0.92	0.046 to 1.83	47
Duration of hospitalization (days)	2 / 218	MD −6.4	−12.6 to −0.1
Time to achieve full feeds (days)	2 / 218	MD −1.8	−2.9 to −0.7

DOI: 10.7717/peerj.2429/table-2

Notes:

MD

Mean difference

RR

Risk ratio

CI

Confidence interval

NEC

Necrotizing enterocolitis

VLBW

Very low birth weight (<1,500 g)

ELBW

Extremely low birth weight (<1,000 g)

VLBW infants

The incidence of severe NEC was significantly reduced in VLBW infants who received probiotics compared to placebo including 25 trials (6,587 patients)—RR 0.47 95% CI [0.36–0.61] (Fig. 5). The incidence of all-cause mortality was significantly reduced in VLBW infants who received probiotics compared to infants who received placebo in 24 trials (6736 patients) with RR 0.74 95% CI [0.61–0.90]. Compared to VLBW infants who received placebo, those who received probiotics had a significantly reduced duration of parenteral nutrition (Table 2).

There was significant heterogeneity in the outcomes of duration of hospitalization, and time to full feeds.

ELBW infants

Eight trials reported outcome data on this weight group (Al-Hosni et al., 2012; Lin et al., 2008; Manzoni et al., 2006; Manzoni et al., 2009; Jacobs et al., 2013; Oncel et al., 2014; Costeloe et al., 2016; Roy et al., 2014). The only trial to enroll infants solely in this weight group was Al-Hosni et al. (2012). ELBW infants were a pre-specified subgroup in the Jacobs trial (Jacobs et al., 2013). In the remaining six trials, outcome data for ELBW infants was presented as a post-hoc subgroup analysis. ELBW infants who received probiotics had a significantly shorter duration of hospitalization and reached full enteral feeding sooner compared to infants who received placebo, see Table 2.

No statistically significant differences were demonstrated for the incidence of NEC (Fig. 6), mortality, culture-proven sepsis, any bacterial sepsis and any fungal sepsis. There was significant heterogeneity in the outcomes of culture-proven sepsis and mortality. Other outcomes were only reported in a small number of patients and trials.

Limitations

The limitations to this systematic review were as follows:

• 1.

  Three of the Chinese language trials (Di & Li, 2010; Deng & Chen, 2010; Zhou, 2012) included in the older review (Yang et al., 2014) could not be obtained in full text and were not included in this review.

• 2.

  No unpublished data was requested from any of the manufacturers of probiotic products assessed in this review.

• 3.

  Only one trial in the previous review addressed long term neurodevelopmental outcomes, but this information could not be confirmed (Kitajima et al., 1997). Akar 2016 (Akar et al., 2016) and the abstract from one of the ProPrems conference presentations (Jacobs et al., 2016) also reports on neurodevelopmental outcomes. If the Kitajima (Kitajima et al., 1997) and ProPrems results were available these could be combined for a summary effect estimate in a future review.

Remaining uncertainties

The outcome of fungal sepsis showed a definite benefit with no heterogeneity (Table 2). Some of the included studies employed antifungal prophylaxis (either systemic or topical) in their infants as per their normal NICU practice. This choice is not the routine practice at all institutions and is not standard practice (Benjamin jr et al., 2014; Austin, Darlow & McGuire, 2013). The impact of these studies with background antifungal therapy was not explored in sensitivity analyses but could be considered in future reviews for its impact on the outcome of fungal sepsis.

Which probiotic product to use remains uncertain, since the total body of evidence comprises a heterogeneous group of probiotics (individual species and combination products, and regimens). In the previous review, only the Lactobacillus and multispecies supplements were shown to be effective for this outcome. We would recommend a regulatory body-approved product and that quality assessment be requested from the manufacturer to validate the purity of product. The evidence of benefit was clear for Lactobacillus or Bifidobacterium species and multiple species products so any of these would be reasonable choices.